Various health communities have expressed concerns regarding whether average bioequivalence (BE) limits (80.00-125.00%) for the 90% confidence interval of the test-to-reference geometric mean ratio are sufficient to ensure therapeutic equivalence between a generic narrow therapeutic index (NTI) drug and its reference listed drug (RLD). Simulations were conducted to investigate the impact of different BE approaches for NTI drugs on study power, including (1) direct tightening of average BE limits and (2) a scaled average BE approach where BE limits are tightened based on the RLD's within-subject variability. Addition of a variability comparison (using a one-tailed F test) increased the difficulty for generic NTIs more variable than their corresponding RLDs to demonstrate bioequivalence. Based on these results, the authors evaluate the fully replicated, 2-sequence, 2-treatment, 4-period crossover study design for NTI drugs where the test product demonstrates BE based on a scaled average bioequivalence criterion and a within-subject variability comparison criterion.